Abstract:Functional Near-Infrared Spectroscopy (fNIRS) is a non-invasive form of Brain-Computer Interface (BCI). It is used for the imaging of brain hemodynamics and has gained popularity due to the certain pros it poses over other similar technologies. The overall functionalities encompass the capture, processing and classification of brain signals. Since hemodynamic responses are contaminated by physiological noises, several methods have been implemented in the past literature to classify the responses in focus from the unwanted ones. However, the methods, thus far does not take into consideration the uncertainty in the data or model parameters. In this paper, we use a Bayesian Neural Network (BNN) to carry out a binary classification on an open-access dataset, consisting of unilateral finger tapping (left- and right-hand finger tapping). A BNN uses Bayesian statistics to assign a probability distribution to the network weights instead of a point estimate. In this way, it takes data and model uncertainty into consideration while carrying out the classification. We used Variational Inference (VI) to train our model. Our model produced an overall classification accuracy of 86.44% over 30 volunteers. We illustrated how the evidence lower bound (ELBO) function of the model converges over iterations. We further illustrated the uncertainty that is inherent during the sampling of the posterior distribution of the weights. We also generated a ROC curve for our BNN classifier using test data from a single volunteer and our model has an AUC score of 0.855.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problem of uncertainty in the classification of functional near - infrared spectroscopy (fNIRS) data. Specifically, fNIRS is a non - invasive brain - computer interface (BCI) used to monitor and image the hemodynamic changes in the brain. However, due to the interference of physiological noise, fNIRS data is usually contaminated, which makes it difficult to accurately classify task - related responses. #### Main problems: 1. **Limitations of existing methods**: Although previous studies have proposed multiple classification methods (such as linear discriminant analysis, support vector machines, etc.), these methods do not consider the uncertainty in the data or model parameters. 2. **Importance of uncertainty**: In fNIRS data, there are two main types of uncertainty: - **Epistemic Uncertainty**: Related to the uncertainty of model parameters and can be reduced by more data. - **Aleatoric Uncertainty**: Related to the noise in the training data set itself and cannot be reduced by obtaining more data. 3. **Lack of models considering uncertainty**: Most of the existing fNIRS classification models ignore these uncertainties, which may lead to inaccurate prediction results in practical applications. #### Solutions: To solve the above problems, the author proposes to use Bayesian neural networks (BNN) for binary classification tasks (right - hand and left - hand finger tapping). BNN uses Bayesian statistical methods to assign probability distributions to network weights instead of point estimates, so it can capture both epistemic uncertainty and aleatoric uncertainty simultaneously. #### Method overview: - **Data processing**: Pre - process the publicly available fNIRS data set, including steps such as band - pass filtering and baseline correction, to remove physiological noise and DC offset. - **Model construction**: Use variational inference (VI) to train the BNN model, and approximate the true posterior distribution by optimizing the evidence lower bound (ELBO) function. - **Performance evaluation**: Evaluate the model performance through the ROC curve and AUC score, and show the convergence of the ELBO function and the change of the posterior weight distribution. Finally, the model achieved an average classification accuracy of 86.44% on the data of 30 volunteers, and the AUC score was 0.855, which proves the importance of considering uncertainty in fNIRS data classification.

Classification of fNIRS Data Under Uncertainty: A Bayesian Neural Network Approach

A Brain-Computer Interface Based on a Few-Channel EEG-fNIRS Bimodal System

Unleashing the potential of fNIRS with machine learning: classification of fine anatomical movements to empower future brain-computer interface

Classification Algorithm for fNIRS-based Brain Signals Using Convolutional Neural Network with Spatiotemporal Feature Extraction Mechanism

A Ternary Bi-Directional LSTM Classification for Brain Activation Pattern Recognition Using fNIRS

Multivariate Kalman filter regression of confounding physiological signals for real-time classification of fNIRS data

Major Surgery in a Subject with Factor V Deficiency

Real-Time Subject-Independent Pattern Classification of Overt and Covert Movements from fNIRS Signals

Feature Extraction and Classification Methods for Hybrid fNIRS-EEG Brain-Computer Interfaces

Transformer Model for Functional Near-Infrared Spectroscopy Classification

Investigation of different approaches for noise reduction in functional near-infrared spectroscopy signals for brain–computer interface applications

Improvement of Classification Accuracy of Four-Class Voluntary-Imagery fNIRS Signals using Convolutional Neural Networks

CNN-based classification of fNIRS signals in motor imagery BCI system

Conditional-GAN Based Data Augmentation for Deep Learning Task Classifier Improvement Using fNIRS Data

Enhancing Classification Accuracy with Integrated Contextual Gate Network: Deep Learning Approach for Functional Near-Infrared Spectroscopy Brain–Computer Interface Application

BayesNetCNN: incorporating uncertainty in neural networks for image-based classification tasks

Classification algorithm for motor imagery fusing CNN and attentional mechanisms based on functional near-infrared spectroscopy brain image

A General and Scalable Vision Framework for Functional Near-Infrared Spectroscopy Classification

Validation of brain-derived signals in near-infrared spectroscopy through multivoxel analysis of concurrent functional magnetic resonance imaging

Cross-Modal Transfer Learning From EEG to Functional Near-Infrared Spectroscopy for Classification Task in Brain-Computer Interface System

Functional Near Infrared Spectroscopy (fNIRS) based Odor Detection and Classification using Functional Data Analysis